perscache 0.3.0

An easy to use decorator for persistent memoization: like `functools.lrucache`, but results can be saved in any format to any storage.

perscache

An easy to use decorator for persistent memoization: like functools.lrucache, but results persist between runs and can be stored in any format to any storage.

• Use cases
• Features
• Getting started
• Make your own serialization and storage back-ends
• API Reference

Use cases

• Cache the results of a function that uses a lot of resources: runs for a long time, consumes a lot of traffic, uses up paid API calls etc.
• Speed up retreival of data that doesn't change often.
• Inspect the results of a decorated function while debugging.

Features

Caching

• Easy to swap out the cache configuration when switching environments.

• Async functions supported.

• Time-to-live (TTL) support - automatically invalidate cache entries after a certain time.

• Automatic cache invalidation when the decorated function arguments or code have been changed.

• You can ignore changes in certain arguments of the decorated function.

Serialization and storage

• Various serialization formats: JSON, YAML, pickle, Parquet, CSV etc.

• Various storage backends:

  • local disk (implemented) or
  • cloud storage (to be implemented soon).

• Serialization and storage are separated into two different classes, so that you can mix various serialization formats and storage back-ends as you like - JSON to local storage, Pickle to AWS, Parquet to Google Cloud Storage etc.

• You can set default serialization format and storage backend and then change them on a per-function basis.

• You can easily add new serialization formats and storage back-ends.

• Local storage is file-based, so you can use human-readable serialization (JSON, YAML, CSV) and inspect cached results.

• Automatic cleanup: least recently used results can be removed from storage when the total storage size exceeds a given threshold.

Getting started

Installation

pip install perscache

Basic usage

from perscache import Cache

cache = Cache()

counter = 0

@cache.cache()
def get_data():
    print("Fetching data...")

    global counter
    counter += 1

    return "abc"

print(get_data())  # the function is called
# Fetching data...
# abc

print(get_data())  # the cache is used
# abc

print(counter)  # the function was called only once
# 1

Changing parameters or the code of the function invalidates the cache

@cache.cache()
def get_data(key):
    print("The function has been called...")
    return key

print(get_data("abc"))  # the function has been called
# The function has been called...
# abc

print(get_data("fgh"))  # the function has been called again
# The function has been called...
# fgh

print(get_data("abc"))  # using the cache
# abc

@cache.cache()
def get_data(key):
    print("This function has been changed...")
    return key

print(get_data("abc"))  # the function has been called again
# This function has been changed...
# abc

Setting the expiry time of the cache

import datetime as dt

@cache.cache(ttl=dt.timedelta(days=1))
def get_data():
    """This function will be cached for 1 day
    and called again after this period expires."""
    ...

Ignoring certain arguments

By specifying the arguments that should be ignored, you can still use the cache even in the values of these arguments have changed. NOTE that the decorated function should be called with ignored arguments specified as keyword arguments.

@cache.cache(ignore=["ignore_this"])
def get_data(key, ignore_this):
    print("The function has been called...")
    return key

print(get_data("abc", ignore_this="ignore_1"))  # the function has been called
# The function has been called...
# abc

# using the cache although the the second argument is different
print(get_data("abc", ignore_this="ignore_2"))
# abc

Changing the default serialization format and storage backend

# set up serialization format and storage backend
cache = Cache(serializer=JSONSerializer(), storage=GCPStorage("bucket"))

...

# change the default serialization format
@cache.cache(serialization=PickleSerializer())
def get_data(key):
    ...

Alternating cache settings depending on the environment

import os

from perscache import Cache, NoCache
from perscache.storage import LocalFileStorage

if os.environ.get["DEBUG"]:
    cache = NoCache()
else:
    cache_location = (
        "gs://bucket/folder"
        if os.environ.get["GOOGLE_PROJECT_NAME"]
        else cache_location = "/tmp/cache"
    )
    cache = LocalFileStorage(location=cache_location)

@cache.cache()
def function():
    ...

Inspecting cached results

When using LocalFileStorage(location=...), the files are put into the directory specified by the location parameter.

The files are named like <function_name>-<hash>.<serializer_extension>, e.g. get_data-9bf10a401d3d785317b2b35bcb5be1f2.json.

Automatic cleanup

When using LocalFileStorage(max_size=...), the least recently used cache entries are automatically removed to keep the total cache size with the max_size limit.

Make your own serialization and storage backends

Although you can use the standard PickleSerializer() for almost any type of data, sometimes you want to inspect the results of a decorated function by lookin into the cache files. This requires the data to be serialized in a human-readable format. But the included human-readable serializers (JSONSerializer(), YAMLSerializer(), CSVSerializer()) sometimes cannot process complex objects.

To see which serializers are compatible with which data types, see the compatibility.py file.

That's when making your own serializer comes in handy.

To do this, you should:

1. Derive your own serialization classe from the abstract Serializer class and override the abstract methods. You should also provide the extension class variable that specifies the file extension.
2. Use your class with the Cache class.

class MySerializer(Serializer):

    extension = "data"

    def dumps(self, data: Any) -> bytes:
        ...

    def loads(self, data: bytes) -> Any:
        ...

cache = Cache(serializer=MySerializer())

Making a custom storage backed is similar:

class MyStorage(Storage):
    def read(self, path, deadline: datetime.datetime) -> bytes:
        """Read the file at the given path and return its contents as bytes.
        If the file does not exist, raise FileNotFoundError. If the file is
        older than the given deadline, raise CacheExpired.
        """
        ...

    def write(self, path, data: bytes) -> None:
        """Write the file at the given path."""
        ...

cache = Cache(storage=MyStorage())

API Reference

class Cache()

Parameters

• serializer (perscache.serializers.Serializer): a serializer class to use for cinverting stored data. Defaults to perscache.serlializers.PickleSerializer.

• storage (perscache.storage.Storage): a storage back-end used to save and load data. Defaults to perscache.storage.LocalFileStorage.

decorator Cache().cache()

Tries to find a cached result of the decorated function in persistent storage. Returns the saved result if it was found, or calls the decorated function and caches its result.

The cache will be invalidated if the function code, its argument values or the cache serializer have been changed.

Arguments

• ignore (Iterable[str]): keyword arguments of the decorated function that will not be used in making the cache key. In other words, changes in these arguments will not invalidate the cache. Defaults to None.

• serializer (perscache.serializers.Serializer): Overrides the default Cache() serializer. Defaults to None.

• storage (perscache.storage.Storage): Overrides the default Cache() storage. Defaults to None.

• ttl (datetime.timedelta): The time-to-live of the cache. If None, the cache never exprires. Defaults to None.

class NoCache()

This class has no parameters. It is useful to alternate cache behaviour depending on the environment.

decorator NoCache().cache()

The underlying function will be called every time the decorated function has been called and no caching will take place.

This decorator will ignore any parameters it has been given.

Serializers

Serializers are imported from the perscache.serializers module.

See also how to make your own serializer.

class perscache.serializers.JSONSerializer

Uses the json module.

class perscache.serializers.YAMLSerializer

Uses the yaml module.

class perscache.serializers.PickleSerializer

Uses the pickle module.

class perscache.serializers.CloudPickleSerializer

Uses the cloudpickle module. It's the most capable serializer of all, able to process most of the data types. It's the default serializer for the Cache class.

class perscache.serializers.CSVSerializer

Uses the pandas module. Processes pandas.DataFrame objects.

class perscache.serializers.ParquetSerializer

Uses the pyarrow module. Processes pandas.DataFrame objects.

Parameters

• compression (str): compression used by pyarrow to save the data. Defaults to "brotli".

Storage back-ends

Storage back-ends are imported from the perscache.serializers module.

See also how to make your own storage back-end.

class perscache.storage.LocalFileStorage

Keeps cache entries in separate files in a file system directory.

This is the default storage class used by Cache.

Parameters

• location (str): a directory to store the cache files. Defaults to ".cache".

• max_size (int): the maximum size for the cache. If set, then, before a new cache entry is written, the future size of the directory is calculated and the least recently used cache entries are removed. If None, the cache size grows indefinitely. Defaults to None.